Reindeer CLI

Use the public reindeer binary for customer-facing Reindeer operations.

SKILL.md · frontmatter

name: reindeer-external-cli
description: >
  Use the public Reindeer CLI to manage workspaces, agents, runs, cases,
  evaluations, connectors, artifacts, contexts, extensions, notifications,
  interactions, feedback, operations, members, and service accounts.
  Trigger for Reindeer CLI setup, workspace administration, agent execution,
  connector operations, case investigation, dynamic UI, or public API discovery.

Operating rules

Run reindeer <resource> --help and reindeer <resource> <verb> --help before
constructing an unfamiliar command. CLI help and the tenant OpenAPI schema are
authoritative; resource schemas evolve.
Pass -w <workspace> for workspace-scoped resources. Names and IDs are both
accepted where help says "ID or name."
Use -o json for programmatic output. List responses include pagination
metadata; follow next_page_token when present.
Prefer full AIP-122 resource names inside JSON bodies, for example
workspaces/<workspace>/agents/<agent>/runs/<run>.
Treat secret values, OAuth client secrets, and artifact download URLs as
sensitive. Do not print or persist them unnecessarily.
Use --wait when available. Otherwise poll the returned operation until
done is true.

Install and authenticate

Install on macOS or Linux:

macOS / Linux

curl -fsSL https://headless.reindeerlabs.ai/install.sh | bash
reindeer --version
reindeer login

Install on Windows:

Windows

irm https://headless.reindeerlabs.ai/install.ps1 | iex
reindeer --version
reindeer login

For a non-interactive agent, use both flags and surface the printed URL to the
user:

shell

reindeer login --headless --no-prompt

Target another tenant or saved profile consistently:

shell

reindeer login --headless --no-prompt \
  --profile demo \
  --base-url https://api.demo.reindeerlabs.ai

reindeer -p demo -b https://api.demo.reindeerlabs.ai -w <workspace> agents list

Useful global flags:

Flag	Purpose
`-w, --workspace <name>`	Workspace name or ID
`-o, --output json\|table`	Output format
`-b, --base-url <url>`	API base URL override
`-p, --profile <name>`	Configuration profile
`-v, --verbose`	Verbose logging

Inspect or diagnose local configuration:

shell

reindeer config list
reindeer config profiles
reindeer doctor

Discover commands and schemas

Start with CLI help:

shell

reindeer --help
reindeer <resource> --help
reindeer <resource> <verb> --help

The public tenant schema is available without authentication:

OpenAPI: https://api.<tenant>.reindeerlabs.ai/api/v1/openapi.json
Swagger UI: https://api.<tenant>.reindeerlabs.ai/api/v1/docs

Use these resource groups:

Resource	Main verbs
`workspaces`	`list`, `get`, `create`, `update`, `delete`
`templates`	`list`, `get`
`connection-profiles`	`list`, `get`, `create`, `update`, `delete`, `test`
`connectors`	`list`, `get`, `create`, `update`, `delete`, `trigger`
`artifacts`	`list`, `get`, `create`, `download`
`contexts`	CRUD, `versions`, `version`, `new-version`
`agents`	`list`, `get`, `create`, `update`, `delete`
`agent-revisions`	`list`, `get`
`evaluations`	`list`, `get`, `create`
`runs`	`list`, `get`, `create`
`steps`	`list`, `get`
`cases`	`list`, `get`, `update`, `delete`, `search`, `aggregate`, `history`
`interactions`	`inbox`, `list`, `get`, `respond`
`feedback`	`list`, `create`
`extensions`	CRUD, `versions`, `version`, `new-version`
`notifications`	`list`, `get`, `create`, `update`, `delete`
`secrets`	`list`, `get`, `create`, `update`, `delete`
`operations`	`list`, `get`
`members`	`list`, `invite`, `update-role`, `remove`
`workspace-members`	`list`, `update`
`service-accounts`	CRUD, `rotate-secret`

Long-running operations

Mutations and interaction responses may return an operation:

shell

reindeer operations get <operation-id> -w <workspace> -o json

A completed operation has done: true and either result or error. An
operation that creates a run completes before the run itself; then poll
reindeer runs get.

Connection profiles and connectors

Connection profiles hold connector authentication. Use the command-specific
help or OpenAPI schema for the selected connection_profile_type.

shell

reindeer connection-profiles create -w <workspace> --data '{
  "connection_profile_id": "<id>",
  "display_name": "<name>",
  "connection_profile_type": "<type>",
  "config": {}
}'
reindeer connection-profiles test <profile> -w <workspace>

Create and trigger an API connector:

shell

reindeer connectors create -w <workspace> --data '{
  "connector_id": "file-upload",
  "display_name": "File Upload",
  "connector_type": "api",
  "config": {"connector_type":"api","trigger_mode":"per_file"}
}'

reindeer connectors trigger <connector> -w <workspace> --file <path>

Pass multiple --file flags for multiple files.

Artifacts

Create a document artifact:

shell

reindeer artifacts create -w <workspace> --file <path> --data '{
  "artifact_type": "document",
  "unique_key": "<dedup-key>",
  "data": "{\"type\":\"document\",\"filename\":\"<filename>\",\"content_type\":\"<mime>\",\"size_bytes\":0}"
}'

data is a JSON-encoded string, not a nested object. Downloading returns a
short-lived signed URL:

shell

reindeer artifacts download <artifact-id> -w <workspace> -o json

Contexts and agents

Before writing, restructuring, or reviewing Context artifacts, read
Writing Effective Context for Reindeer Agents.
Use it to separate system-prompt controls, operating policy, schemas, and
reference data; preserve one canonical owner per rule; and review for narrative,
slop, and conflicting rules. For retrieval-only or routine CLI operations, the
commands below are sufficient.

Context entries reference artifacts by full resource name:

shell

reindeer contexts create -w <workspace> \
  --name <name> \
  --display-name "<display name>" \
  --entries '[{"artifact":"workspaces/<workspace>/artifacts/<artifact-id>","display_name":"<entry>","description":"<description>"}]'

reindeer contexts new-version <context> -w <workspace> \
  --entries '[{"artifact":"workspaces/<workspace>/artifacts/<artifact-id>","display_name":"<entry>","description":"<description>"}]'

A new context version can trigger compilation of a new agent revision.

Create an agent with a connector subscription and context:

shell

reindeer agents create -w <workspace> \
  --name <name> \
  --display-name "<display name>" \
  --subscriptions '{"connector_type":"api","connector":"workspaces/<workspace>/connectors/<connector>","events":[{"type":"api_trigger"}]}' \
  --contexts workspaces/<workspace>/contexts/<context>

Pass each subscription as a separate --subscriptions value.

Runs, steps, and cases

Create a run with --input. The required type is new_run,
previous_run, or previous_case.

shell

reindeer runs create -w <workspace> --agent <agent> --wait \
  --input '{"type":"new_run","user_input":"<instruction>","artifacts":[]}'

Resume a run session:

shell

reindeer runs create -w <workspace> --agent <agent> --input '{
  "type":"previous_run",
  "previous_run":"workspaces/<workspace>/agents/<agent>/runs/<run>",
  "session":{"session":"resume","user_input":"<follow-up>"}
}'

For previous_run and previous_case, session is a nested object. Use
"session":"new" for a fresh session linked to the prior input. Poll the run
until its status is no longer running.

Inspect related execution data:

shell

reindeer steps list -w <workspace> --agent <agent> --run-id <run>
reindeer cases list -w <workspace> --agent <agent>
reindeer cases search -w <workspace> --agent <agent> --text "<text>"
reindeer cases aggregate -w <workspace> --agent <agent> --group-by <field>
reindeer cases history <case> -w <workspace> --agent <agent>

Check each case subcommand's help for filters, pagination, and aggregation
options.

Evaluations

Evaluate a revision against its latest golden dataset, or pass full case
resource names explicitly. Evaluation commands require CLI 0.3.32 or newer.

shell

reindeer evaluations list -w <workspace> --agent <agent>
reindeer evaluations list -w <workspace> --agent <agent> --revision <revision>
reindeer evaluations get -w <workspace> --agent <agent> <evaluation-resource-name>
reindeer evaluations create -w <workspace> --agent <agent> \
  --revision <revision> --wait
reindeer evaluations create -w <workspace> --agent <agent> \
  --revision <revision> \
  --cases workspaces/<workspace>/agents/<agent>/cases/<case-1> \
          workspaces/<workspace>/agents/<agent>/cases/<case-2>

create always requires --revision. For get, a full evaluation resource
name carries its revision; with a bare evaluation ID, also pass --revision.
Use get for per-case scores and the criteria rollup; list is a summary.

Interactions and feedback

Find pending human-in-the-loop requests and respond:

shell

reindeer interactions inbox -w <workspace> --agent <agent> --status pending
reindeer interactions respond <interaction> -w <workspace> \
  --agent <agent> --run-id <run> --text "<answer>"
reindeer interactions respond <interaction> -w <workspace> \
  --agent <agent> --run-id <run> --decision allow --note "<optional note>"

Use --selected <label...> for option labels. A 409 means the interaction
is no longer pending.

Create textual feedback for a run, step, or case:

shell

reindeer feedback create -w <workspace> --agent <agent> --data '{
  "feedback_context":{"type":"run_feedback","run":"workspaces/<workspace>/agents/<agent>/runs/<run>"},
  "feedback_data":{"type":"textual","sentiment":"positive","text":"<feedback>"}
}'

Notifications

Notifications deliver platform events through a connection profile:

shell

reindeer notifications create -w <workspace> --wait \
  --name run-done \
  --display-name "Run complete" \
  --events run_completion \
  --connection-profile <profile> \
  --scope '{"type":"workspace"}' \
  --enabled true

Use update --enabled false to pause delivery.

Extensions and dynamic UI

Extensions are versioned display surfaces. Treat ExtensionType,
ExtensionPayload, and the dynamic UI component schema in the tenant OpenAPI
document as authoritative.

shell

reindeer extensions create -w <workspace> --wait \
  --name <name> \
  --type <type> \
  --parent workspaces/<workspace>/agents/<agent> \
  --display-name "<display name>" \
  --payload '{"type":"<type>","view_spec":{}}'

reindeer extensions new-version <extension> -w <workspace> \
  --payload '{"type":"<type>","view_spec":{}}'

update changes mutable metadata; new-version replaces the served payload.

For display_case_decision and display_case_decision_outlook, render data is
{case, artifacts: {input}}. Read case state from /case/state/..., read input
artifact references from /artifacts/input, and keep editable write paths
relative to case.state (for example, read /case/state/status but write
/status). DocumentPreview.fieldsPath must bind to the resolved fields array,
not a plain path string. Do not add a top-level view_spec.state.

Secrets, members, and service accounts

Secret values are write-only:

shell

reindeer secrets create -w <workspace> \
  --name <name> --description "<description>" --value "<value>"

Organization and workspace access:

shell

reindeer members list
reindeer members invite --email <email> --org-role <role>
reindeer workspace-members list -w <workspace>
reindeer workspace-members update <member-id> -w <workspace> --role <role>

Service accounts are organization-scoped OAuth clients:

shell

reindeer service-accounts create \
  --display-name "<name>" \
  --org-role organization_viewer \
  --workspace-roles '{"<workspace>":"workspace_editor"}'
reindeer service-accounts rotate-secret <service-account>

Create and rotate return the client secret once. Capture it securely; list and
get never return it.

Keep the CLI and skill current

shell

reindeer upgrade
reindeer install-skill

install-skill offers detected Claude Code, Codex, and GitHub Copilot
installations, lets the user choose user or project scope, and delegates skill
placement to gh skill install. For a non-interactive update, use:

shell

reindeer install-skill --agent <claude|codex|copilot> --scope user --yes